Twitch Data Scientist Interview Guide

Twitch is a live video streaming platform that allows users to watch and broadcast live-streamed or pre-recorded videos of the broadcaster’s video game gameplay. The platform is owned and operated by Twitch Interactive, a subsidiary of Amazon. Founded in 2011 as an offspring of the “stream anything platform”, Justin.tv, its prime focus is on streaming live video games, including broadcasting Twitch’s own hosted eSports competitions. Besides these functions, Twitch also broadcasts music and creative content, which can be viewed live on the site or from “Video on Demand”.

As of Q4 of 2019, Twitch has over 3.6 million average monthly broadcasters, just shy of figures above 3.8 million from March 2020. With over 56 thousand concurrent Twitch broadcasters and 1.44 million concurrent viewers on average, Twitch’s web APIs handle over 50 thousand requests a second, which translates to over 2.3 billion hours watched in 2019.

With the amount of data generated from different daily streams, Twitch’s data science team performs a wide range of analysis to help shape product decisions. This feat is achieved through enhanced data pipelines “that collects data, cleans data, and loads over a billion events per day into their data warehouse.”

Twitch has a “science team”, consisting of titles and roles related to data science. It is supported by three pillars, namely “data science research, user experience research, and data governance.” Data science sits right in the middle of these three science organizations and collaborates with the other teams on many occasions.

Data scientists’ roles at Twitch are greatly influenced by the teams they are working with, and, as such, the roles and functions may range from product-focused analytics to machine learning and deep learning algorithms. Currently, there are two main types of data scientists: the “product strategy-oriented data scientists”, who provide business-impact insights from data analysis, and the “data product data scientists”, who build specific algorithms and techniques yield new products informed by data.

Required Skills

Twitch only hires qualified data scientists with a minimum of 3 years (5 years plus for senior data scientist roles) industry experience in data science-related projects.

Note: applications are processed and evaluated based on specific industry experience related to the job roles on the teams.

Other relevant qualifications include:

• Advanced degree (MS or Ph.D.) in Computer Science, Mathematics, Statistics, or related fields.
• Experience in applying supervised and unsupervised algorithms to large-scale data.
• Sound background in data-science fundamentals: data manipulation in R or Python, SQL, and statistics (Hypothesis testing, Regression, etc.).
• Advanced knowledge of machine-learning techniques applied to large-scale data.
• Experience with building data pipelines, data warehouses, dimensional modelling, building aggregates, and optimizing data workstreams from data preparation to analysis to deployment.
• Sound background in AWS.
• Experience with database management system software, e.g., MySQL, PostgreSQL, etc.
• Experience designing and assessing the impact of A/B experiments in a product development cycle.
• High level of comfort with creating dashboards in Tableau or similar software.
• Experience in data analysis and communication around data, including experimentation, data visualization, and defining KPI strategy for business.
• Experience with programming languages like R, Python, C/C++, Go, or Java.
• Product analytics experience with signup funnels, engagement metrics, and retention analysis.
• Proficiency in SQL and ETL/ELT in a business environment with complex data sets.

Although Twitch has a dedicated “Science Team” consisting of data scientists, data analysts, and data engineers, data scientists are often embedded within other teams and sometimes collaborate with other departments. As a large company with data scientists working in over 20 teams, on the individual level, roles at Twitch are inherently tied to specific teams.

Based on the team’s needs, data scientist roles at Twitch may include:

• Advertising Product: leveraging a long list of advanced data analytics tools and methodology to work on a wide range of challenging problems including econometric modelling and auction dynamics, pricing and segmentation, and maximizing value for viewers and advertisers.
• Core Product: setting up and tracking KPIs, designing experiments, evaluating A/B tests, and implementing data instrumentation for developing strategy and evaluating/improving product plans.
• Alliances: defining and tracking KPIs, supporting strategic initiatives, evaluating new lines of business, and helping shape the way performance is measured with Twitch’s deals.
• Financial Planning and Analysis: overseeing data instrumentation, design dashboard/report building, and metrics reviewing to guide financial decisions and provide business-impact insights.
• Marketing: leading analysis and optimization of event marketing programs. Working cross-functionally with go-to-market roles, including growth, product, content events, and brand marketing teams.
• Mobile: developing strategy and evaluating product initiatives within the Mobile team through defining KPIs, designing experiments, evaluating A/B testings, and supporting strategic initiatives.
• Core Science: driving evidence-based decision-making throughout Twitch with data analytics and machine learning models. Collaborating with specialists in data science, analytics, engineering, and economics disciplines to effectively develop reliable and reproducible analyses at scale.

Initial Screen

Twitch’s initial data scientist screen interview is a 30 to 60-minute phone chat with a hiring manager, discussing the team, Twitch as a community, your ideas on data, your technical background, and how your past relevant projects and experiences align with the job roles on the team.

Sample Questions:

• If you had all possible data and unlimited leeway, what’s the first change you would make to Twitch’s site?
• What’s one non-academic thing you’ve done that you’re proud of?
• What is your biggest weakness?

Technical Screen

Twitch’s data scientist technical screen is very similar to most tech companies. This interview involves a one-hour live screening on a coderpad with a data scientist, and the questions asked are usually SQL-based. There is also an element of behavioral and background experience in this interview.

Onsite Interview

The onsite interview is the last stage in Twitch’s data scientist interview process. This process comprises 6 one-on-one interviews (or conferences) split around behavioral, experimental, SQL, and coding questions with a product manager, data scientists, technical product manager, and analyst.

Each interview is approximately 45 minutes long, and the questions on these interviews tend toward experimentation, A/B testing, business intelligence, and heavy analytics. There are also product-focused and behavioral rounds, with questions around business analytics experience, past working experience, and your knowledge of Twitch’s culture. Take a look at our guide to data science case studies for practice.

Tips for Twitch Interviews

Twitch’s Data Scientist interview is a combination of data science concepts standardized to assess an applicant’s ability to apply statistical and analytics concepts to understanding and predicting user behavior and answering business questions based on the analysis.

It helps to brush up on your knowledge of:

• statistics and probability
• time series analysis
• experimental designs
• A/B testing
• predictive modelling concepts

It also helps to know the metrics used at Twitch, especially those related to products and features.

Twitch offers an ecosystem that allows employees to thrive and be the best version of themselves by encouraging them to get their hands dirty and find something they love. A lot of emphasis is placed on building high-performing teams through mentorship programs, and in fact, the ability and desire to mentor is something Twitch looks for in applicants.

• Create a histogram/bin this data using SQL.
• What metrics would you track in a particular A/B test?
• What metric/s would you optimize for? How would you pick the “winner” of the A/B test?
• How would you create bins?

See more Twitch data science questions from Interview Query: